Andre – Python, Big Data, Data Science

The app is an internal Machine Learning system to block and prevent financial fraud. The company has a diverse range of products that serve as payment methods for third-party services and products.

The main functionality of the app is to have a 24/7 online endpoint that analyses batch data and searches for fraudulent patterns. Those analyses foster the analytics team to block clients that are committing or attempting to commit financial fraud. This helps the company avoid millions of reais in fraud and high-risk third-party business.

• Created ML pipelines from data acquisition to model deployment.
• Gathered, Processed, and Stored data;
• Deployed ML models;
• Versionized Code, Data, and ML artifacts;
• Drew and implemented the back-end architecture;
• Implemented unit and integration tests;
• Created Dockerized back-end microservices;
• Refactored and redesigned not performative code;
• Led Data Scientist team to follow SW development best practices.

The project was a Web App designed to solve computer vision tasks, such as new client onboarding via scanned documents, OCR, Image Classification, Image Segmentation, and Video Analytics. The main target audience was Fintechs, Banks, and companies that had Video Monitoring without automation. The app had a front-end interface where the users could submit pictures and videos which they wanted to analyze. There was also an API for those who needed to automate the pipeline or process batch data. The back-end part was built on GCP/AWS and was hosted in Kubernetes Clusters. The app was adopted by a big Brazilian bank and by a big Latin American fintech.

• Created ML pipelines from data acquisition to model deployment.
• Gathered, Processed, and Stored data;
• Train, Test, and Validate Computer Vision models;
• Versionized Code, Data, and ML artifacts;
• Drew and implemented the back-end architecture;
• Implemented unit and integration tests;
• Created Dockerized back-end microservices;
• Refactored and redesigned non-performative code.

The App was developed as an internal sub-component of another Desktop Software called Visual Components. The latter is used to create a Digital Twin of factories which uses automation and robots. The target audience was an Industrial Robotic company called JPM Industry. The main feature of the app was to automatically program a path for the robotic arm to act, avoiding collisions, given only the starting and ending points. This could expedite the programming process by days and was a key resource to adapt to constant changes in palletization applications.

• Developed robotic arm path planning algorithm;
• Refactored and maintained legacy code;
• Designed and built front-end for user interaction with the app;
• Wrote unit tests for backend components;
• Wrote and published four papers;
• Won best-paper award at the ICARSC 2019 conference.